Machine learning topological defects in confluent tissues

Area of Science:

• Physics and Biology
• Emerging paradigms in biological systems characterization

Background:

• Active nematics are crucial for understanding biological systems.
• Defects in active nematics play a key role in biological processes.
• Existing defect detection methods are unsuitable for non-rod-shaped cells, like epithelial layers.

Purpose of the Study:

• To develop a convolutional neural network (CNN) for detecting and classifying nematic defects in confluent cell layers.
• To create a method applicable to experimental images of cell layers, particularly those with non-rod-shaped cells.

Main Methods:

• Development of a convolutional neural network (CNN).
• Training the CNN on experimental images of cell layers.
• Demonstration of defect detection on experimental data with non-rod-shaped cells.

Main Results:

• The CNN successfully detects and classifies nematic defects in confluent cell layers.
• The developed method is suitable for cells that are not rod-shaped.
• The machine learning model outperforms current defect detection techniques.

Conclusions:

• The new CNN method significantly improves the accuracy of experimental data interpretation for nematic defects.
• This approach reduces the data required for accurate defect property capture.
• The findings advance the study of nematic defects in biological systems, offering cost and accuracy benefits.